An algorithmic feed may finally become a choice instead of a condition of entry. Australia has proposed making social platforms ask users whether they want personalized recommendations or a feed limited to accounts they follow. That is the sharpest development in this edition of AI news today—and a useful reminder that AI policy increasingly reaches product design, not just model training.

The rest of the briefing moves from policy to deployment. Google is taking AI-guided contrail avoidance into a larger airline trial and backing 16 environmental projects across Asia-Pacific. OpenAI is funding practical newsroom experiments in Ukraine. In Britain, a prominent AI-policy architect is leaving a government research post after accepting a role at Anthropic. And Travis Kalanick’s new company is reportedly rebuilding an autonomous-vehicle team with $100 million from Uber. The common thread is control: who chooses what the system optimizes, who oversees it, and who gets to switch direction.

AI News Today: Australia Proposes an Off Switch for Algorithmic Feeds

Australia’s government released draft Digital Duty of Care legislation for targeted consultation. Its “My Feed, My Way” proposal would require social platforms to notify new and existing users that they can choose a personalized feed or opt out and see content from friends and creators they follow. The draft also extends safety duties to games, apps and AI chatbots used by minors. Non-compliance could attract penalties of up to A$109.2 million.

This is a proposal, not enacted law, and the government says it will seek feedback before introducing legislation later this year. Still, the mechanism is unusually concrete: it targets the default that gives recommender systems their leverage. For platforms, implementation details will decide whether the choice is meaningful or hidden behind friction. For everyone else, it shifts the policy question from “are algorithms harmful?” to “should a person be able to decline the optimization?” Europe’s transparency rules, covered in our EU AI Act compliance guide, are part of the same move toward visible, testable controls.

Google and Cathay Pacific Put Contrail Avoidance Into Live Operations

Google Research and Cathay Pacific are expanding an operational trial that uses AI predictions, satellite imagery and weather data to identify airspace where persistent contrails are likely to form. Dispatchers and pilots can then plan small altitude changes within normal safety procedures. Google says more than 80 flights followed suggested routes in an initial phase and that its satellite analysis estimated a roughly 40% reduction in contrail warming impact.

That 40% figure is a company estimate from a limited trial, not a settled industry-wide result. The importance is operational: this is AI influencing real flight planning with existing aircraft, not generating another sustainability dashboard. The next phase across Asia-Pacific and transpacific routes should reveal how well the approach handles different weather, traffic and fuel constraints. If the benefit survives at scale, modest route changes could become a practical climate tool while slower aircraft and fuel transitions continue.

Google Backs 16 Environmental AI Teams Across Asia-Pacific

A separate Google DeepMind initiative selected 16 organizations for a three-month AI for the Planet accelerator. The cohort ranges from bioacoustic wildlife monitoring and crop guidance to mangrove mapping, carbon verification and urban-energy optimization. Participants receive access to Google’s AI stack, technical support and mentorship.

Accelerators are inputs, not outcomes; selection does not prove that a project will scale or deliver measurable environmental gains. What makes this cohort worth watching is its concentration on verification and field data. Climate AI often fails at the handoff between a good model and a usable local system. The meaningful evidence will be hectares monitored, pests caught earlier, carbon claims checked and operating costs low enough for communities to keep using the tools after the program ends.

Anthropic Hire Triggers a Governance Reset at Britain’s ARIA

Matt Clifford said he will step down as chair of the UK’s Advanced Research and Invention Agency after accepting a full-time role leading Anthropic’s government engagement outside the United States. The Guardian reports that senior MPs had raised conflict-of-interest concerns. Clifford plans to remain until November 6 with safeguards in place.

No evidence in the report says Clifford improperly influenced a decision. The issue is institutional confidence: a person who helped shape British AI policy and chaired a publicly funded research agency is moving directly into a frontier lab’s government-relations operation. Recusal rules can address individual decisions, but not every information advantage or appearance of influence. Watch for the government’s account of its conflict assessment and for tougher cooling-off or disclosure rules around the increasingly busy revolving door between AI labs and public policy.

OpenAI Takes Newsroom AI Training Into Ukraine

OpenAI, the World Association of News Publishers and the Association of Independent Regional Press Publishers of Ukraine announced a program to help Ukrainian news organizations test AI in editorial, audience, product and commercial work. The Newsroom AI Catalyst will give ten publishers hands-on support, implementation roadmaps and API credits.

This is a partnership announcement, not evidence that AI has already strengthened newsroom resilience. It does, however, focus on the unglamorous middle of adoption: selecting a useful case, changing a workflow and measuring whether the result saves time or earns revenue without weakening editorial standards. In a conflict environment, verification, security and human accountability cannot be afterthoughts. The best outcome would be repeatable local practices rather than dependency on donated credits or a single vendor.

Uber’s $100 Million Atoms Bet Reopens the Robotaxi Loop

The Financial Times reports that Uber invested $100 million in Atoms, Travis Kalanick’s industrial-AI company, and that Atoms has assembled autonomous-vehicle talent and held early talks about technology for Uber’s network. The report follows Atoms’ earlier $1.7 billion funding round. Atoms told the FT it has no plans to enter the “saturated robotaxi market,” while adding that Uber may use its technology for ridesharing if useful.

The distinction matters. Atoms could sell autonomy software without operating a fleet or consumer service. The $100 million figure and robotaxi discussions are reported facts from unnamed sources, not a disclosed commercial contract. For Uber, the logic is familiar: spread bets across autonomy providers while keeping the marketplace. For Atoms, the next proof point is a named platform, safety case and deployment—not a hiring list that makes Silicon Valley’s history rhyme rather loudly.

Watch & Learn

Editor’s note: Google Cloud Tech’s Recommendation Systems Overview is a compact primer on what recommenders do, why companies value them and where they go wrong. Give it about ten minutes. It is useful for product leaders and curious users who want enough technical grounding to understand what Australia is actually proposing people should be allowed to switch off.

AI, Translated: Recommender System

A recommender system predicts which items you are most likely to engage with, then ranks what appears next. A video app might combine your watch history with patterns from similar users, score thousands of candidates and show the top few. It does not simply “know your taste”; it optimizes a chosen goal using imperfect signals. You should care because the goal—watch time, purchases, satisfaction or diversity—quietly shapes the information and choices placed in front of you.

Try This Today: Audit a Policy Against Your Real Workflow in Gemini

Goal: Turn one product policy into a five-minute decision checklist. Gemini Apps can analyze uploaded documents; Google’s official file-upload guide notes that access and limits vary by account, and Workspace administrators may need to enable Drive access.

1. Download the privacy or AI-use policy for a tool your team uses.
2. Upload it to Gemini and describe one real workflow, including the data involved.
3. Ask for citations to the document and verify every “allowed” conclusion yourself.

Copy-ready prompt: “Compare this policy with our workflow: [describe task, users and data]. Create a table with policy clause, permitted action, prohibited action, unresolved ambiguity and safer alternative. Cite the page or section for every row. Do not infer permission from silence; write ‘not stated’ instead. Finish with three questions we should send to the vendor or our legal team.”

One Thing to Remember

AI control is becoming a product feature, a governance rule and an operating discipline at the same time. The most important switch is often the one that lets a person decline the system’s preferred optimization.